slm benchmark index : [https://es-d-43788997920260729-019fa301-76b4-7fb1-9566-ef7608be22d5.codepen.dev/]

Dumb-1.2-RC5

This may be the last "RC".

benchmark reuslt

==================================================
               EVALUATION SUMMARY               
==================================================
 ARC-Easy Accuracy : 35.94%
 BLiMP Accuracy    : 70.40%
--------------------------------------------------
 Wikitext Byte PPL : 3.187
 Wikitext Word PPL : 500.6
==================================================

Effort

We were attracted by the idea of the Glint-2's effort control and created an Effort that can work with a more versatile model.

Low Effort Generated Continuation: "Once upon a time in a world where there is not anything else that was the culprit to what we would have thought about as we are, and our future being brought back to us. We must have been looking for something more like an experiment with how things can be used as an asset when"

Generation Speed : 21.79 tokens/sec

Medium Effort Generated Continuation: "Once upon a time in a world where there was no space for space, the U.S. government has been fighting against the United States since the 1960s. The US Department of Education (DOE) recently released its first report, The New York Times: A Brief History of the American World, which includes 30 new books and articles about the Great Britain's Great Britain. It is not until 2015 that the news has fallen to prominence, but it doesn't have to be as"

Generation Speed : 7.41 tokens/sec

High Effort Generated Continuation: "Once upon a time in a world where the demand for energy is rapidly increasing, it’s impossible to get there. In fact, one way of thinking about how we can do that is by looking at how our money works and why they are important to us. When we"

Generation Speed : 22.60 tokens/sec

Downloads last month
193
Safetensors
Model size
34.6M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support